superlite
flowchart LR
A(["You: say what you want"]) --> B["superlite writes a plan"]
B --> C{"You approve"}
C --> D["Agents build in parallel"]
D --> E["Review and fix"]
E --> F(["Report: time and cost"])
You decide twice: what to build, and whether the plan is right. Everything after that is automatic. If you would rather not be asked, pick the AI orchestrator and it decides for you.
What a run looks like
Section titled “What a run looks like”This site was built by superlite. Here is that run in five screens.
-
You say what you want.
Claude Code · superlite ❯ Build a documentation site for superlite, laid out like Warp's docs.superlite suggests a mode, asks who orchestrates, and opens a branch of its own for the run, so another chat in the same project cannot overwrite it.
-
You read a plan, not code.
Claude Code docs/plans/2026-11-12-docs-site.review.mdReply approve to start, or tell me what to change.The plan is plain language with diagrams and a rough time and cost. Nothing runs until you answer. You decide: approve it.
-
You approve and agents build.
Claude Code ❯ looks good, go aheadA plan of 12 tasks went out. Up to six agents worked side by side, reviewers checked the work and tests ran along the way.
-
You read the report.
Claude Code · build report Projected Actualwall-clock (min) 6.8 6.3agent cost (USD, +rework) 1.83 2.66The forecast was fixed before the build, so the report shows honestly where the run missed it. Nothing is committed yet.
-
You ship it.
/superlite:finishruns the tests and offers merge or PR. Nothing is pushed without your yes.The pull request shows the merged result, with the summary written from the plan.
The full walkthrough, with what happens underneath, is on the Quickstart.